Aborted propagation of the Ethiopian rift caused by linkage with the Kenyan rift

Continental rift systems form by propagation of isolated rift segments that interact, and eventually evolve into continuous zones of deformation. This process impacts many aspects of rifting including rift morphology at breakup, and eventual ocean-ridge segmentation. Yet, rift segment growth and interaction remain enigmatic. Here we present geological data from the poorly documented Ririba rift (South Ethiopia) that reveals how two major sectors of the East African rift, the Kenyan and Ethiopian rifts, interact. We show that the Ririba rift formed from the southward propagation of the Ethiopian rift during the Pliocene but this propagation was short-lived and aborted close to the Pliocene-Pleistocene boundary. Seismicity data support the abandonment of laterally offset, overlapping tips of the Ethiopian and Kenyan rifts. Integration with new numerical models indicates that rift abandonment resulted from progressive focusing of the tectonic and magmatic activity into an oblique, throughgoing rift zone of near pure extension directly connecting the rift sectors.


Supplementary Note 1: Tectonics of the Turkana depression
The Turkana depression is a portion of the East African rift, a >3000km long aligned succession of adjacent individual tectonic basins resulting from the ongoing separation between the African and Somalian plates 1 . Plate divergence is occurring in a roughly E-W direction, at rates of a few millimetres per year 1 : the velocity of plate motion decreases from the northern Ethiopian rift (~5 mm yr -1 ) southwards; extension at the latitude of Lake Turkana occurs at rates of ~3 mm yr -1 (ref. 1). The Turkana depression is the region of interaction between the Ethiopian and the Kenyan rifts. In both rifts, extensional deformation gives rise to a typical narrow rift morphology, with large fault escarpments (with vertical displacement of >1 km) separating a roughly 80-100 km wide rift valley from uplifted plateaus. Conversely, faulting, earthquakes and Quaternary-Holocene volcanic centres in the Turkana depression are widespread over a width of more than 300 km ( Supplementary Fig. 1). In addition, the Turkana region lacks the typical rift valley morphology dominated by large fault escarpments. Instead, extension results in numerous faults with limited vertical offset clustered within three main different sub-parallel domains of deformation 2 : (i) the Lake Turkana basin, (ii) the Kino Sogo fault belt and (iii) the Ririba rift (see main text). The anomalous structural pattern and physiography of the Turkana depression is likely controlled by variations in the pre-rift structure of the lithosphere 3 . Geophysical data 4,5 provide evidence for significant crustal thinning beneath the NW-SE-trending Turkana depression. This thinning resulted from Mesozoic-Paleogene rift events, which created a system of NW-SE grabens (such as the Anza graben in Ethiopia, or the South Sudan rifts) 6 at a high angle to the Cenozoic Ethiopian/Kenyan rift valleys ( Supplementary Fig. 1). The inherited thin crust related to these Mesozoic-Paleogene events was suggested to cause the low topography of the region with respect to the surrounding uplifted plateaus 4 . Numerical and physical modelling of rifting indicate that the transversal pre-existing heterogeneity with thinned crust and lithosphere also likely controlled the transition from localized deformation in the rift valley domains to the diffuse deformation in the Turkana depression, therefore influencing rift linkage in the region 3 . Similarly, pre-existing inherited structures have been suggested to have largely controlled the pattern of faulting at the scale of individual faults and fault arrays 7,8 . The Oligocene-Recent evolution, summarized in Supplementary Fig. 1, is characterised by the early activation of different N-S trending basins (e.g., Lotikipi, Gatome, Lokichar) West of Lake Turkana 9,10,11,12,13,14,15 (Supplementary Fig. 1). These basins started developing during the Oligocene-Early Miocene, after the earliest manifestation of flood volcanism at ∼45 Ma in South Ethiopia 2 . The activity of the westernmost basins (Lotikipi, Gatome) ceased during the Middle Miocene. During the Early-Middle Miocene, extension started in some of the basins related to the Kenyan rift system (e.g., Samburu, Turkana, Kerio), and in the Chew Bahir basin. This activity was contemporaneous with a first, minor phase of extensional deformation in the Southern Main Ethiopian rift ( Supplementary Fig. 1). Major phases of extension started in the Late Miocene in the Gofa Province of South Ethiopia and in the Southern Main Ethiopian rift (Abaya, Chamo, Galana basins and Amaro Horst region). During the Pliocene-Quaternary, the volcano-tectonic activity focused in the Lake Turkana basin, now a direct continuation of the Kenya rift, whereas the extensional deformation related to the Ethiopian rift have been suggested to have propagated southward into the Eastern Turkana depression to form the Ririba rift and the Hurri Hills and Marsabit volcanic lineaments ( Supplementary Fig. 2). Overall, a progressive eastward migration of Oligocene-Recent deformation within the Turkana depression (from early extension in the basins West of Lake Turkana to the domains of active deformation of Lake Turkana, Chew Bahir-Kino Sogo and Ririba), has been suggested 2 ( Supplementary Fig. 2). This migration, together with a northward propagation of rifting in the Kenya rift and a southward propagation of the Main Ethiopian rift to the Ririba and Marsabit volcanic lineaments, has been considered to have contributed to the anomalous breadth of the zone of deformation in the region 2 .

Supplementary Note 2: Analysis of maximum displacement (D)/length (L) profiles and influence of pre-existing basement structures on faulting
We have analysed the maximum displacement (D) and trace length (L) relationship for the major normal faults that bound the Ririba rift on its western margin ( Supplementary Fig. 3). The D/L relations varies over different length scales depending on several factors. These include: (1) material property, (2) type of fault including position and direction of measurement, (3) earthquake rupture and slip/propagation history, (4) segmentation and linkage (evolution), and (5) reactivation of pre-existing structures (see ref. 16 and references therein). The D/L relations can be used to characterize the growth model of faults, which can be referred to two end-members: (1) the segment growth and linkage model, where initially isolated fault segments elongate and accumulate displacement via simple tip propagation before eventually linking to form a through-going fault system, and (2) the constant length model, in which fault lengths are rapidly established and, thereafter, growth is dominated by displacement accumulation with limited (if any) lateral propagation 17 . This latter growth model is directly applicable to reactivated fault systems in which fault lengths are inherited from underlying structures and established almost instantaneously on geological timescales 17 ; the earliest phases of growth of reactivated faults are therefore characterized by a significant under-displaced nature. D-L curves from the boundary faults of the Ririba rift ( Supplementary Fig. 4) are characterized by flat top profiles, low-end gradients and low D-L values, which are in agreement with the constant length model. This supports our hypothesis that the development of major faults in the Ririba rift is controlled by reactivation of old, large basement structures. This is also confirmed by analysis of basement fabrics in the region, which show three main trends, oriented respectively NNW-SSE, NW-SE and NE-SW ( Supplementary Fig. 5). The NNW-SSE trend is well expressed in the surroundings of the Chew Bahir basin 7 , where it has been shown to have exerted a strong control on the development of the major normal faults bounding the depression 18,19 . The NW-SE basement fabric parallels the orientation of the Mesozoic Anza graben and South Sudan rifts and diffusely characterise the area East of the Ririba rift and the Mega escarpment. The NE-SW fabric is represented by minor structures diffused throughout the region of interest and has been associated to the existence of a large, 30 km-wide transversal fault zone (Buluk fault zone 7,8 ) marking the southern termination of the Chew Bahir and Segen basins ( Supplementary Fig. 5). The different sets of faults characterizing the Ririba rift are sub-parallel to these basement fabrics ( Supplementary Fig. 5). In particular, the main boundary faults are characterized by a striking similarity in orientation with the NNW-SSE fabric; similarly, the minor sets of faults giving rise to the complex angular patterns parallel the NNE-SSW, NE-SW or ENE-WSW basement trends. This parallelism, the highly variable orientations of the Ririba rift faults, which are typically not orthogonal to the extension direction, and the above observations on the D-L curves support a strong influence of the pre-existing structures on fault development, as also observed in the nearby Kino Sogo fault belt 7,8 .

Supplementary Note 3: Chemical analysis of samples of volcanic rocks
Whole rock analyses were performed on 9 samples from the Dilo area, collected both from the Pliocene lava flows clearly cut by the main Ririba faults and from the products discordantly lying on these lava flows or directly sealing the faults (Supplementary Table 2). Results of this analysis show that samples from the faulted Pliocene lava pavement of the Ririba rift (MDZ4, 5 and 12) plot in the field of subalkaline basalts ( Supplementary Fig. 8), while the products from the top of the sequence, and in some cases sealing the faults of the Ririba rift, plot in the fields of basanite/tephrite up to foidite and are similar in composition to the Quaternary products of the area (e.g., ref. 4). The minor compositional variability observed in the Quaternary products is possibly partially related to the presence of abundant (locally up 10-15% by volume) mantle xenoliths or xenocrysts, and to processes of fractional crystallization.
Zr/Nb ratios (Supplementary Fig. 9) confirm the presence of two compositionally distinct groups, with the Quaternary lavas showing a nearly constant low ratio of 3.2, close to the value of 3.5 suggested for primary magmas of basanite-phonolite trend of the area 20 . Conversely, Pliocene rocks show a higher Zr/Nb ratio (8)(9) comparable to that observed from continental tholeiites 21 .

Supplementary Note 4: Analysis of vent alignment
The results of the analysis of vent alignment ( Supplementary Fig. 10) indicate an overall, NE-SW elongation of the Dilo-Dukana volcanic field, with a complex distribution of the vents at a more local scale. The main peaks of vent distribution are indeed at N10°-20°E, N40°-50°E, with a minor peak in distribution at N20°W. Overall, these results indicate that the majority of the vent alignments are not orthogonal to the regional, current E-W direction of extension and therefore do not directly respond to stresses controlled by plate motion. Instead, the overall elongation of the field, subparallel to a major pre-existing structure (the Buluk fault zone) hypothesized west of the Ririba rift 7,8 (see above Supplementary Note 2), is suggestive of a strong control exerted by inherited basement fabrics on the distribution of volcanic vents. This is supported by distribution of vents at local scale, which is controlled by the angular network of NNE-SSW, NE-SW or ENE-WSW structures that reactivate preexisting fabrics, as discussed above.

Supplementary Note 5: Other geological and morphological evidence for fault inactivity and constraints on the timing of rift abandonment
Many different lines of evidence indicate a lack of Pleistocene deformation activity in the Ririba rift and surrounding areas. Apart from the Late Pleistocene large lava flows burying the faults, recent fault inactivity is confirmed by the presence on the two main faults of the rift of small drainages that locally incise the scarps and generate 100-200 m-long fans ( Fig.  3; Supplementary Fig. 11). These small fans, made up of colluvium, have their apex upstream the fault scarp indicating that the sedimentary rate is higher than the slip rate of the faults. As a further confirmation, elsewhere the Ririba River channel is located in the middle of the half graben, not because the sediment input from the footwall is enough to fill the site of maximum subsidence 22 , but because the river is not forced by the subsidence of the hangingwall to flow close the fault plane. Analysis of satellite images indicates that the normal faults at the southern termination of the rift affect the Late Miocene-Pliocene Bulal lavas but are covered by the basal flows of the Hurri Hills edifice (Supplementary Fig. 12a). Similarly, faults of the complex, angular network affecting the basal lava platform South-West of the Mega escarpment are sealed by Quaternary lava flows erupted from vents at the Northeastern termination of the Hurri Hills ( Supplementary Fig. 12b). A few minor normal faults affecting basal Hurri Hills lavas (see area indicated with arrow in Supplementary Fig. 12) suggest that very limited extensional deformation in the area may have occurred during the initial phases of building of this volcano. The only published data on the Hurri Hills activity are the K/Ar ages by Brotzu et al. 10 and those reported in the geological map of the North Horr 23,24 (see also ref. 25). According to these ages and to the geological map, the present edifice is formed by a Late Pliocene basal lava pile, forming a large shield volcano, overlain by the products (lava flows and tephra blankets) of a large number of Pleistocene monogenetic vents (mainly cinder cones) located on the shield surface. The basal flood lavas of the edifice started emplacing at around 3 Ma, with a sample from the southwestern slope of the edifice giving a poorly constrained K/Ar age of 2.8±1.3 Ma (refs. 23,24; Supplementary Figure 7). This timing is consistent with these lavas overlying the Bulal basalts, which we dated at 3.6-3.7 Ma in the area of Dilo. Conversely, the lava flows and tephra products of the following monogenetic activity span quite a large time period, from an oldest age of 2.3±0.17 Ma (K/Ar age of ref. 10) up to a mid-Pleistocene age (K/Ar age of 0.54±0.10 Ma reported in refs. 23,24) for some products of summit vents. As stated above, the lavas of the Hurri Hills shield seal the Ririba faults. Therefore, since the basal lavas of this shield, whose oldest reported age is 2.8 Ma, are only weakly affected by few minor normal faults at its northern termination and they are overlain by undeformed younger products with ages as old as 2.3 Ma, the end of rift activity likely occurred close to the Pliocene-Pleistocene boundary. Of course, the possibility exists that the lavas covering the faults may be even older than the poorly constrained age of 2.8 Ma (e.g., the ca. 3 Ma age reported in refs. 23,24), therefore further reducing the duration of faulting in the Ririba rift. Future work is needed to decrease the uncertainty in the timing of rift deactivation.

Supplementary Note 6: Analysis of the Mega escarpment
In order to provide additional constraints on the evolution and distribution of deformation in the area, we have analysed the morphostructural evidences for recent or current activity along the Mega escarpment, which represents the Eastern margin of the Turkana depression. The escarpment exposes high grade metamorphic rocks (mostly gneiss) of the Precambrian basement, locally covered by recent basaltic lava flows, scoria cones and maars of the Mega-Megado volcanic field. The ~50 km long escarpments, separating the depression (elevations <900 m asl) from the Mega plateau (elevations >1400 m asl) and reaching a maximum height of >1200 m, is separated into two main segments by a fluvial valley ( Supplementary Fig. 13). Apart from three ephemeral rivers that interrupt it, most of the escarpment is incised by small drainages whose channels rarely breach into the plateau. The escarpment is strongly eroded so that it is possible just locally to recognize few triangular facets. To get a quantitative indication on the activity of the tectonic structure that generated the escarpment, we measured the mountain front sinuosity following ref. 26:

=
where Lmf is the mountain-piedmont junction and L is the overall length of the mountain front. The resulting value of 1.7 is included in Class 2, interpreted as typical of a low tectonic activity 26 .
At the bottom of the escarpment there is a pediment dipping to the SW and covered by many fans close to the mountain front. The fans appear to be inactive: the very few active channels are debris flow-like. Probably they occur during the main rainy season (April and May) when it rains ~53% of the scarce annual precipitation (~100 mm yr -1 ; refs. 27,28). Although the fans are inactive, it is possible to recognize three different types, named F1, F2, F3 ( Supplementary Fig. 13). F1 type, located in correspondence with the present three ephemeral rivers cutting across the Mega escarpment, presents very low slope (1-3°), is large in shape (~10-12 km wide) and could include more than one fan, some of which have the apex far from the mountain front in a telescopic pattern. The second type (F2) has a low slope (3-5°) and forms a sort of belt (~5 km wide) at the bottom of the Mega escarpment in correspondence with channels draining just the scarp. The relationship between F1 and F2 landforms indicates a contemporary deposition. On top of F2 lies the third type (F3), that include small (~1 km wide) and steep (5-8°) fans locally incised by active debris flow-like channels. The configuration of the fans suggests that in the past, during a phase of wetter climate (likely the African Humid period, 15-5 ka), rivers draining the Mega Plateau were able to generate wide fans: their telescopic pattern is usually associated with rivers characterized by an erosion rate higher than the deformation rate at the mountain front 29,30,31 , therefore supporting the interpretation based on the mountain front sinuosity. In the present semi-arid conditions, the steepness of the small fans (F3 type) at the bottom of the escarpment, as well as the active short channels, are more typical of a debris flow dynamic. Notably, to the South of the study area, two very recent lava flows belonging to the volcanic field of Mega-Megado follow the channels along the escarpment from the high standing plateau down to the lowlands with no displacement. According to these considerations, the fault that generated the Mega escarpment is inactive or has an activity so low that the surface processes easily overcome any possible displacement.

Supplementary Note 7: Numerical models
The complete evolution of rifting in the numerical models is illustrated in Supplementary Fig.  14. As explained in the main text, model results clearly illustrate the localization of deformation in the two offset rift valleys, which interact in correspondence to the transversal heterogeneity. During the initial phases of extension (12-24 km of bulk extension, roughly corresponding to 12-9 m.y. before present), deformation within the transversal domain is diffused over a wide region, in line with what is observed in previous modelling 3 . For increasing bulk extension (36 km of bulk extension, corresponding to 6 m.y. ago), the region affected by tectonic activity becomes narrower, focusing in the two overlapping tips of the propagating rift valleys (see labels TE and TK in Supplementary Fig. 14). In the latest stages of the model (60 km of bulk extension, corresponding to the present day), the two rift valleys directly connect, cutting away the southern tip of the Ethiopian rift valley (i.e., the Ririba rift in nature) from active deformation.  Rybacki et al. 40 Hirth & Kohlstedt 41 Hirth & Kohlstedt 41